The selective nature of visual attention prioritizes objects in a scene that are most perceptually salient, thoserelevant to personal goals, and animate objects. Here we present data from two intentional change detectionstudies designed to determine the extent to which animals in a scene distract from other changes. Our stimulidepicted camouflaged animals in their natural habitats. We compared participants' responses to changinganimals and inanimate objects selected from the samepictures, thus improving on othermethodologies studyingthis effect. Experiment 1 results suggest that animals are noticed rapidly and accurately, even when they sharebottom-up features with the rest of the scene. Additionally, the unchanging presence of camouflaged animalsdistract from detecting inanimate changes. Experiment 2 employed signal detection theory (SDT) to measurethe sensitivity (d′) and response bias (β) related to changing animate versus inanimate stimuli. Experiment 2outcomes indicate that participants tend to adopt a liberal response bias and are most sensitive to animatechanges. Presence of an animal in a scene also influences sensitivity (d′) when participants had to attend toand notice inanimate changes. Our findings are interpreted as additional support for the animate-monitoringhypothesis which suggests that early detection of animacy may have endowed our hunter-gather ancestorswith survival advantages.
Research on Atypical Develop-
ent of Psychological Sciences,niversity of California, Merced,
Attention is often studied as a domain-general mechanism.Perspectives in evolutionary psychology deviate from this interpretation,suggesting that attention is better understood as a set of specialized sys-tems designed to solve particular adaptive problems (Tooby & Cosmides,2005). The animatemonitoring hypothesis, proposed by New, Cosmides,and Tooby (2007), suggests that domain-specific mechanisms prioritizeand monitor animate stimuli in the environment. A logical consequenceof allocating attention to monitor animate stimuli is that inanimatestimuli become less salient when animals are present. The followingpaper presents the use of a unique methodological approach designedto determine if animate stimuli are continually monitored by thehuman attention systems. We explore how perceptual sensitivities andbiases toward the animate hamper the detection of inanimate objectsin the presence of animate distractors.
1.1. Theory and background
Within the workings of the visual attention system, not all elementsof a visual scene are attended to equally (Simons & Levin, 1997). Whenobjects are incongruent, or do not fit into the surrounding landscape(e.g. a fire hydrant in the living room) people notice the odd-ball objectsfaster and with greater accuracy (Hollingworth & Henderson, 2000).When something near the ‘interesting’ portion of a scene-changes, wefocus our attention on those details first (Rensink, O'Regan, & Clark,1997). Similarly, greater experience in a particular domain leads tofast detection of familiar objects as they change (Werner & Thies, 2000).
Animate features of a scene are given higher priority duringattention tasks, suggesting that making distinctions between animateand inanimate stimuli is particularly relevant to human cognition. Inits rudimentary form, this ability appears almost immediately afterbirth (see Opfer & Gelman, 2011) and serves as one of the foundationsfor social and cognitive development. Human and nonhuman animalshave agency (Spelke, Phillips, & Woodward, 1995) and provide sociallyrelevant information (Baron-Cohen, 1995). Brain architecture alsoreflects the importance of our capacity for distinguishing animatefrom inanimate information. Dedicated domain-specific neuralnetworks govern this ability and their disruption is associated withdebilitating deficits in verbal expression and concept formation(Caramazza & Shelton, 1998). Distinguishing animate from inanimate
2 M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
is so critical that it is one of the longest-preserved cognitive functionsin patients with neuro-degenerative disorders (Hodges, Graham, &Patterson, 1995).
Caramazza and Shelton (1998) proposed that these animacy-honedbrain networks evolved in response to selection pressures encounteredby our species. Animate stimuli are directly relevant to survival (Tooby& Cosmides, 2005), and contain perceptual features that have remainedconsistent throughout hominid evolution. Human and non-humananimals have specific characteristics (e.g. eyes, faces, fur, teeth, claws),and create recognizable stereotypical motion that quickly capturesattention (Cavanagh, Labianca, & Thornton, 2001; Cutting & Kozlowski,1977; Pratt, Radulescu, Guo, & Abrams, 2010).
Animate agents can be food or foe. As such, it is probable thatwe evolved capacities for perceiving predators and prey in ourenvironment (Barrett, 2005). Participants viewing arrays of humanfaces orient to facial features associated with threat and anger(Hansen & Hansen, 1988; Lundqvist & Ohman, 2005). Preferentialattention is given to fear inducing non-human stimuli (e.g. snakes andspiders) compared to inanimate objects (Lipp, Derakshan, Waters,& Logies, 2004; Ohman, Flykt, & Esteves, 2001). Taken together, evidencesuggests that stimuli relevant to threat are prioritized in visualsearch tasks.
Early and accurate detection of nonthreatening animate agents alsolikely provided increased opportunity tomate or cooperate. Preferentialattention is directed to attractive distractor faces, which interfereswith completion of other tasks (Sui & Liu, 2009). Even with limitedinformation, familiar biological motion can be identified as belongingto a friend (Cutting & Kozlowski, 1977) or an animal (Pavlova,Kragloh-Mann, Sokolov, & Birbaumer, 2001). Broadly, perspectives inevolutionary psychology argue for the presence of domain-specificsystems that process animals with mnemonic, attentional, and learningcircuitry that operates concordantly depending on task demands(Tooby & Cosmides, 2005).
New et al. (2007) hypothesized that the human attention systemevolved category-specific selection criteria to prioritize and frequentlymonitor animate stimuli. Using a modified version of the flicker-taskparadigm (Rensink et al., 1997), the authors demonstrated how ourattention systems preferentially detect animals and humans in visualscenes. When compared to familiar inanimate objects (e.g., coffeemugs and telephones) changes to animals and people were noticedfaster and with greater accuracy. This effect remained even whenanimate changes were very small and when vehicles (i.e., objects thatare equally familiar, spatially/temporally sensitive, and potentiallydangerous, but not part of our ancestral past) were included amongthe comparisons.
The animate monitoring hypothesis, as proposed by New et al.(2007), also predicts that attention systems actively monitor animateinformation in a visual scene. Thus, it stands to reason that prioritizingand keeping track of the animate also leads to interference with theprocessing of inanimate information. The stimuli and methodologyused by New et al. (2007), however, are not sufficient for testing thiscomplimentary research question.
New et al. (2007, 2010) presented participants with 70 uniqueimages and asked them to detect visual changes in the pictures.Scene-changes were evenly divided into 5 categories, correspondingto the levels of the independent variable (i.e., scenes with appearing/disappearing animals, people, fixed objects, movable objects, andplants), but presence of animate distractorswas not controlled, infusinga potential confound. One or more non-target animate objects weredepicted in 64% of scenes portraying inanimate movable-objectchanges, in 28% of scenes portraying inanimate plant changes, and in50% of scenes portraying inanimate fixed-object changes. Some ofthese animate distractors were quite prominent and centrally locatedin the scenes. Also, 14% of scenes illustrating animate human changesand 42% of scenes illustrating animate non-human changes showedanimate non-targets.
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
This uneven distribution of animate non-targets across the fiveexperimental conditions presents a threat to internal validity.New et al. (2007, 2010) went to great lengths when equating their 70scenes to control for stimulus salience, color, luminance, and contrast,but the content within each scene varied. When comparisons weremade, they were ultimately comparisons of participants' responses todifferent images.
We agree with New et al.'s (2007) animate monitoring hypothesisand have designed our experiments to test a logical follow-up question.If animate stimuli receive attentional priority and are continually mon-itored,will inanimate change-detection behampered by the presence ofanimate objects? This paper presents a cleaner methodology looking atwhether presence of animate distractors in a scene will interfere withparticipants' ability to detect inanimate change.
2. Experiment 1
Experiment 1was designed to further examine the cognitive processthat assigns preferential attention to animate features of a changingvisual scene. This study presents a refinement of the Flicker Task usedby New et al. (2007, 2010). The aim was to extend prior findings bycontrolling for previously unaccounted for confounds including thepotential impact of animate distractors. Experiment 1 serves as amore stringent investigation of humans' apparent attentional bias foranimate objects.
We hypothesized that scene-changes would be noticed with greaterspeed and accuracywhen animate objects changed in a scene. Secondly,we predicted a reduction in participants' ability to notice inanimatescene-changes while animate objects were visible. We hypothesizedthat animals would act as distractors in situations where inanimatechanges had to be detected. We expected attention to prioritizeanimate objects, even when that animate object was not the target.Subsequently; detection of inanimate objects would be deprioritized,impacting response time and accuracy, any time an animal was alsovisible in the scene.
Lastly, we hypothesized that the physical distance (cm) between theanimate and inanimate objects would be related to the ease of changedetection. Flickering inanimate objects should be noticed more rapidly,and with a higher degree of accuracy, the closer their proximity to theanimate distractors. If human cognition prioritizes animacy, andanimals are attended to first, thenwe expected it should be easier to de-tect targets (i.e., inanimate) thatwere closest to the animate non-target.
2.1. Method
2.1.1. ParticipantsParticipants were self-referred college students who volunteered as
part of fulfilling their undergraduate course requirements. All studentswith normal or corrected-to-normal vision, and no prior history ofattention difficulties, were offered a chance to participate. Trainedundergraduate research assistants completed informed consent anddebriefing procedures with all participants.
Thirty-six undergraduate student volunteers were recruited tofacilitate researchers' standardization of task stimuli. Data collectedfrom these 36 participants were not used or considered part of theactual experiment.
One-hundred and five (N = 105) undergraduate students(73 F) were recruited to participate in the actual study with a meanage of 19.04 years.
2.1.2. Materials and apparatusResearchers selected thirty-seven digital pictures from a Google
image search, using the following search terms: “camouflaged animals”and “animals hiding in plain sight.” Permission to use these pictures inexperimental paradigms was obtained from all copyright holders.Three additional pictures were selected from the collection used by
eference for animacy impacts change detection, Evolution and Human
3M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
New et al. (2007, 2010), for a total of 40 pictures. Because perception ofthreatening stimuli attracts attention, carewas taken to exclude animalsassociated with fear processing (e.g. snakes and spiders).
After this preliminary selection phase, researchers screened thepictures to ensure equivalence on perceptual properties. A detaileddescription of the process used to screen, select, and equate all visualstimuli on key perceptual properties is included in Appendix A.
After undergoing screening and selection, 33 photographs wereretained from the original collection of 40 images (see Table 1). Threeversions of each scenewere created. One version had the animate targetremoved from the scene. A second version had the inanimate targetremoved from the scene. The third version had both the animate and in-animate targets removed from the scene. In total, 99 scene-variationswere generated.
As shown in Fig. 1, thirty-three ‘Animal Flicker’ stimuli were pro-duced, showing the rapid alternation between an original scene, awhite screen mask, and the same scene with the animal removed.Thirty-three ‘Inanimate Flicker with Animal’ stimuli were generatedshowing a quick switch between the original scene, a white screenmask, and the same scene with the inanimate object removed. The finalthirty-three ‘Inanimate Flicker w/out Animal’ stimuli quickly alternated be-tween pictures with the animate target removed, a white screen mask,and the same scene with both animate and inanimate targets removed.
In summary, our modifications to the flicker paradigm facilitatedincreased control of scene content across experimental conditions.Additionally this methodology allowed us to vary one feature withineach scene while keeping all other aspects constant.
Table 1Visual qualities of the photograph scenes used for flicker stimuli.
Scenenumber
Animate target Inanim
Stimulus Height at tallestpoint (cm)
Width at widestpoint (cm)
Luminance(cd/m2)
Stimul
1 Monkey (C) 5.00 10.00 55.79 Log (C2 Boar (P) 4.00 4.50 77.11 Tree (P3 Chinchilla (M) 3.00 6.50 68.04 Rock (4 Owl (P) 18.00 6.00 71.62 Tree b5 Wolf (P) 25.50 6.00 34.02 Tree (P6 Deer (C) 3.25 4.00 99.65 Log (P7 Bob cat (C) 5.50 6.50 151.24 Bush (8 Wolf (C) 11.00 11.00 77.86 Tree (P9 Leopard (C) 7.50 5.00 135.61 Rock (10 Owl (P) 6.00 2.50 156.7 Tree (P11 Duck (C) 13.50 4.00 106.89 Patch o12 Monkey (C) 2.50 2.00 89.69 Tree b13 Falcon (C) 5.00 9.00 101.52 Rock (14 Bird (C) 4.00 3.50 60.03 Leaves15 Green parrot (P) 6.00 2.50 71.83 Leaf (P16 Tree squirrel(C) 5.00 6.50 78.45 Grass (17 Monkey (P) 5.00 4.00 37.09 Snowy18 Weasel (M) 5.00 10.50 197.84 Pine co19 Deer (P) 6.50 2.50 112.4 Dry gr20 Lion (P) 4.50 2.50 81.09 Tree b21 Bear (C) 0.75 2.00 73.69 Tree (C22 Hippo (C) 3.00 3.00 74.86 Grey r23 Giraffe (P) 22.00 3.00 89.44 Tree (P24 Wolf (P) 7.50 1.50 113.63 Tree (P25 Billy goat (P) 7.00 11.50 125.01 Rock (26 Squirrel (P) 9.00 3.50 99.33 Bush (27 Parrott (P) 2.50 4.00 80.16 Tree b28 Crocodile (C) 1.00 9.50 111.65 Rock (29 Elephant (P) 1.50 2.50 93.82 Tree (C30 Chipmunk (P) 2.50 5.00 99.38 Rock (31 Duck (C) 6.00 7.00 132.7 Bush (32 Stork (C) 4.00 0.75 103.88 Rock (33 Leopard (M) 2.75 4.00 121.22 Tree st
Note: Central position of target, within the scene is indicated by (C).Peripheral position of the target, within the scene is indicated by (P).Targets falling on the dividing lines between center and periphery, within the scene are indicaLuminance values were obtained from Adobe Photoshop Creative Suite–Version 6.Experiment 2 did not use photographs 31, 32, or 33.
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
Graphic modification of visual stimuli was performed using AdobePhotoshop Creative Suite - Version 6. The E-Prime software package(Psychology Tools Software Inc., 2012) was used to administer ourflicker task to participants, on Dell computers, with standard QWERTYkeyboards and 25” diagonal LCD displays.
2.1.3. ProcedureThis experiment was conducted as a single-group, randomized, one-
way design. Stimulus type served as the independent factor, comprisingthree levels, corresponding to the three variations of flicker stimuli(i.e., animal flicker, inanimate flicker with animal, and inanimate flickerw/out animal).
At the start of the experiment, participants were seatedapproximately 61 cm in front of a high definition computer monitor.They were presented with on-screen instructions and given a chanceto ask questions or clarify the task demands. Participants self-startedthe experimental flicker task. The experiment was conducted in aquiet distraction-free environment and the task took approximately15 minutes to complete.
To capture participants' ability to detect visual changes in a scene,our flicker task presented each volunteer with 33 flicker stimuli.Stimuli were randomly selected, without replacement, guaranteeingthat every participant was exposed to each of the 33 originallyretained photographs. Using a counterbalancing strategy, everyoneviewed 11 randomly selected flicker stimuli from each of the threestimulus-type categories.
ate target Distance b/wanimate andinanimate targets
Fig. 1. Examples of the experimental flicker stimuli, with white screen masks in between, alternating presentations of changed and unchanged scene-variations.
4 M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
Participants saw a fixation cross in the middle of the screen for 500milliseconds (ms), followed by the presentation of each scene-pairflick-er stimulus (see Fig. 1). The first image of each scene-pair was displayedfor 250ms, followed by awhite screen for 250ms. The alternate versionof the image was then displayed for 250ms, followed by a white screenfor 250 ms. Each flicker stimulus repeated this sequence for 60 secondsor until the participant detected the change.
Participants were instructed to respond, by pressing the spacebar, assoon as they saw the target appearing and disappearing in the scene.Immediately after responding, participants were shown the corre-sponding photograph with the target object visible. Volunteers wereinstructed to use the computer's mouse to indicate the location of theappearing and disappearing target. Once position of the target objectwas specified, a centered fixation cross appeared followed by presenta-tion of the next flicker.
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
Speed of change detection was the domain of primary interest. Re-sponse times (RTs) were recorded inmilliseconds reflecting the latencybetween the moment each scene-pair first appeared on the screen andthe instant participants pressed the spacebar.
Accuracy of change detection was the domain of secondary interest.Consistent with procedures in New et al. (2007, 2010) a ‘hit’ wasdefined as a mouse-click response with the pointer positioned within1 cm of each target's nearest boundary. The response was coded as a‘miss’ if participants did not detect the change within 60 seconds or ifparticipants' accuracy did not reflect a ‘hit’.
Proximity of inanimate targets to non-target animals was examinedas amoderating variable. Distance in centimeters, between the center ofeach target and non-target, was measured independently and con-firmed by two senior researchers (see Table 1). A proximity index wascomputed and utilized only for instances where the inanimate object
eference for animacy impacts change detection, Evolution and Human
5M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
appeared and disappeared, while the animal remained unchanging inthe scene.
2.1.4. AnalysesOne-way analysis of variance (ANOVA) tests were conducted to ex-
amine differences in participants' RTs and accuracy rates. The indepen-dent fixed-factor comprised three levels, distinguishing the variation inflicker stimulus-type viewed by participants [i.e., (1) animal flicker:scene-pairs where an animal appeared and disappeared, (2) inanimateflicker with animal: scene-pairs where the animal was always visible,while an inanimate object appeared anddisappeared, and (3) inanimateflicker without animal: scene-pairs where the animal was never visible,while an inanimate object appeared and disappeared].
Post-hoc paired comparisons were performed using two-tailedTukey's honestly significant difference (HSD) tests. One-tailed correla-tional analyses were carried out to examine the relationship betweenaccuracy, response times, and the proximity of inanimate targets tonon-target animals in the scene. All statistical analyses were performedusing SPSS 22.0 (IBM Corp, 2013).
2.2. Results
2.2.1. Preliminary data screening and aggregationPrecursory examination of raw data revealed significant positive
skew in the distribution of change detection RTs. Log-transformationwas used to convert response time data to approximate a normal distri-bution. RT conversion was performed independently, for each of the 99flicker stimuli.
All outliers were removed prior to log-transformation. Raw RTs fall-ing outside the 95th percentile (i.e., standardized Z-statistic = ±1.96)were treated as outliers and were excluded. Responses correspondingto RTs b500 ms were also removed before data conversion. These ex-ceedingly quick RTswere considered invalid responses because changescould only be detected after the first flicker, and not before the first500 ms had passed. Approximately 9.4% of participants' original re-sponses were withdrawn from further analysis (animal flicker: 3.2%, in-animate flicker: 3.4%, and inanimate flicker w/out animal: 2.8%)
Relevant summary statistics for log- and non-transformed responsetimes (RTs) are provided in Table 2. All inferential statistics and post-hoc testswere conducted using only log-transformed data [transformedRTs and SDs are reported as (RTlog ± SDlog ms)].1
The remaining log-transformed datawere combined to derive compos-ite average scores. For each of the one-hundred and five participants, threeRTlog aggregate scoreswere calculated, reflecting each participant's averageresponse speed to each of the threeflicker stimulus-types. Composite accu-racy rateswere also computed for eachof the 99flicker stimuli, to representthe proportion (%) of participants who correctly identified the location ofeach appearing/disappearing target. In total; three-hundred fourteen RTlogaverages (n = 314) and ninety-nine accuracy percentage composites(n=99)were created for examinationwith ANOVA. One RTlog composite,aggregating one participant's Inanimate Flicker with Animal score, couldnot be generated due to exclusion of extreme values.
Notably, any RTs that coincided with inaccurate reports of targetlocation (i.e., participants indicated seeing a scene-change more than1 cm outside a target object's boundary) were excluded from allaggregations and analyses. All data summaries (Table 2), as well as allANOVA and post-hoc test results were generated using only responsesthat correctly identified a scene change (i.e., ‘hits’).
2.2.2. Principal analysesOne-way analysis of variance (ANOVA) findings yielded a statistical-
ly significant main effect for stimulus type (F2,311 = 43.21, p b 0.001;
1 Exploratory ANOVA and post-hoc statistics were initially derived using non-transformed data, yielding statistically significant findings. However, due to extremeskew, researchers decided to use log-transformed data for all analyses.
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
η2=0.217). As shown in Fig. 2, changes in a scenewere detected fastestwhen the flickering objects were animals (log-transformed RTlog =3.54 ± 0.16 ms). Participants were slower to detect changes wheninanimate objects were appearing and disappearing in scenes wherethe animal was visible (RTlog = 3.78 ± 0.21 ms) and in scenes wherethe animal was not visible (RTlog = 3.72 ± 0.21 ms).
ANOVA results also reflected significant differences in participants'accuracy rates when comparing the three variations of scene-pair stim-uli (F2,96 = 6.413, p b 0.003; η2 = 0.118). Noticeably fewer ‘hits’ oc-curred when participants had to pinpoint the position of a flickeringinanimate object. Fig. 2 illustrates the low accuracy rates for locating in-animate targets, irrespective of the animal's presence (percent hit =67.33 ± 22.46) or absence (percent hit = 68.50 ± 23.53) from thescene. In contrast; when the flickering target was an animal, partici-pants were more often correct when identifying its precise location(percent hits: 84.36 ± 18.41).
Post-hoc tests confirmed that scene-changes were detected moreslowly, and with less accuracy, when the flickering targets were inani-mate objects (see Table 3). Also as predicted, animals appeared to func-tion as ‘distractors’ in scenes where participants had to detect theappearance and disappearance of inanimate targets. Post-hoc testsshowed that flickering inanimate objects were noticed significantlyfaster in scenes where animals were not visible, as compared to sceneswhere the animals were visible (two-tailed Tukey HSD; p b 0.048).
Correlational analyses found that higher accuracy rates (RTlog: r =−0.379, p b 0.001) were predictably associated with decreasing dis-tance between animal non-targets and flickering inanimate targets.Having an animal ‘distractor’ in the scene did not, however, make par-ticipants any less accurate when detecting changes in inanimate targets(p b 0.837). A modest trend towards quicker reaction times (r= 0.052,p b 0.079) was also observed.
2.3. Discussion
Experiment 1 built on previous investigations demonstrating atten-tional prioritizing for animacy. Initial studies investigating this effect(New et al., 2007) used 70 natural scenes.Within each picture an objectfrom one of five categories (two animate and three inanimate) was se-lected as the target. Methodologically, this created a potential threat tointernal validity as comparisons were made not only between levels ofthe IV, but also between fundamentally different stimuli. Our experi-ment was created to control for such potential confounds.
Additionally, this experiment was designed to address a second re-search question. Within each category of IV, the presence of uncon-trolled animate distractors may have influenced detection of changingtarget stimuli. We hypothesized that the presence of animate objectswithin the scene would distract participants from detecting the changeto the inanimate objects.
Experiment 1 findings indicate that patterns in selective attentionfor animal-related stimuli were uniquely different, when compared topatterns in selective attention for inanimate objects. Changes in thescene were detected faster and accuracy errors were less likely whenthe appearing/disappearing targets were animals. Supporting our sec-ond hypothesis, animals also served as ‘distractors’ when they werepresent in a scenewhere the changing target was inanimate, thus inter-fering with a participant's ability to detect a changing inanimate object.When an animal was visible, it took substantially more time to see theappearance/disappearance of inanimate stimuli. Further, as the distancebetween visible animal ‘distractors’ and inanimate targets increasedparticipants' response speed slowed. They also were less accuratewhen pinpointing the inanimate targets' position.
3. Experiment 2
Experiment 2was conducted to frame the effects observed in exper-iment 1, and byNewet al. (2007, 2010), from the viewpoint of the signal
eference for animacy impacts change detection, Evolution and Human
Table 2Descriptive data for the experiment 1 sample.
Condition Scene characteristic Numberof scenes
Medianpre-log RT(ms)
Pre-logRT SE(ms)
Pre-logRTskew
Post-logRTlogskew
Average% of hits
SE for %of hits
Scene type Animal flicker Animal appears and disappears,inanimate object visible in the scene
33 2973.00 247.27 2.88 0.68 84.36 3.20
Inanimate flickerw/out animal
Inanimate object appears and disappears,animal not visible in the scene
33 5168.00 376.96 1.92 0.17 68.50 4.10
Inanimate flicker with animal Inanimate object appears and disappears,animal visible in the scene
33 6201.00 386.91 1.72 0.02 67.33 3.91
6 M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
detection theory (SDT). This approach offers specific advantages for un-derstanding the perceptual and behavioral processes that help peopledistinguish animate from inanimate changes in selective attentiontasks (Stanislaw & Todorov, 1999). More specifically, experiment 2allowed the exploration of differences in people's perceptual sensitivityto animate changes. It also was intended to give greater insight into theperceptual and/or behavioral response bias that accompanies animatescene-changes.
SDT's strength is its ability to measure the degree to which visualstimuli are perceptually registered (i.e., sensitivity), while simulta-neouslymeasuring the decision process (i.e., bias) that yields responsesto those stimuli (Palmer, Verghese, & Pavel, 2000). SDT methodologydispenses with overly-complex explanations, handles confounds relat-ed to pre- versus post-attentional processes, and discounts the limita-tions of higher-order semantic storage (Verghese, 2001; Wilken & Ma,2004). Use of SDT has the potential to distinguish change blindness asa failure to perceptually organize visual scenes into component partsfrom change blindness as a failure to encode and/or compare objectrepresentations in pre- and post-change scenes (McAnally et al.,2010). In short, experiment 2 incorporates the consideration of falsealarm responses in addition to hit accuracy and response speed.This improves the empirical accounting of how and why animatechanges are detected or missed, in the presence of distractors or ‘noise’.
Designing experiment 2 as an SDT study allows evaluation of the se-lective animacy bias as a perceptual phenomenon.Witt, Taylor, Sugovic,and Wixted (2015) describe standards for using sensitivity andresponse-bias metrics for interpreting SDT outcomes. Particular atten-tion is given to distinguishingbias driven by internal response strategies(e.g., tendency to say ‘yes’ or ‘no’) from bias driven by the kind ofperceptual effects that create illusions (Morgan, Hole, & Glennerster,1990). Employing an SDTdesignmakes it possible to control for internaldecision-based biases like the speed-accuracy tradeoff (Johnson &Proctor, 2004). Finally, it helps make a stronger argument for aperception-driven, animacy-selective change detection system.
Fig. 2. Differences in response time and accuracy
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
3.1. Study rationale
From the perspective of SDT, an individual's sensitivity is his or herability to detect the presence of a signal in a noisy environment (Abdi,2007). The greater an individual's sensitivity, the better he or she is atdiscriminating between the strength of the signal and the strength ofsurrounding noise. The signal seems more salient to the more sensitiveindividual (Stanislaw & Todorov, 1999).
When we overlay the principles of SDT onto our adaptation of theflicker task paradigm, a highly relevant translation of sensitivityemerges. D-prime becomes the measure of the capacity to distinguishbetween an instance where a visual change occurs in a scene (signal),and an instance where there is no visual change (noise). Thus, greatersensitivity can be understood as increased preferential attending tothe appearance/disappearance of any object within a scene becausethe change is more salient to the observer. Scenes where no visualchanges occur are deprioritized as noise.
Within the SDT framework, response bias or beta is the cognitivedecision point where a person has enough information to indicatewhether a signal is present or not (Johnson & Proctor, 2004). A “liberal”decision requires less information before concluding that a signal wasdetected, resulting in lower β values. A more conservative bias is indi-cated by a higher β value, reflecting a need formore information to con-clude that a signal is presentwithin the noise (Pastore & Scheirer, 1974).
Experiment 2was designed to facilitate themeasurement of key SDTindices, the derivation of sensitivity (d′) and response bias (β). The pri-mary aimwas to identify differences in sensitivity to animate versus in-animate changes in scene and examine attentional response bias withrespect to animate and inanimate scene-changes. The impact of animatedistractors on humans' sensitivity and response bias for inanimatescene-changes was also evaluated.
To help disentangle perceptual from decision-based effects,experiment 2 also followed the recommendations of Witt et al. (2015)in creating true signal-present and signal-absent stimuli. As such,
between the three experimental conditions.
eference for animacy impacts change detection, Evolution and Human
Table 3Post-hoc two-tailed Tukey HSD comparisons for experiments 1 and 2.
Experiment 1 Experiment 2
Contrasts Response time (ms) Accuracy rate (% hits) Sensitivity (d′) Response Bias (β)Animal flickervs.inanimate flicker with animal
p b 0.001 p b 0.005 p b 0.001 p b 0.009
Animal Flickervs.inanimate flicker w/out animal
p b 0.001 p b 0.01 p b 0.001 p b 0.002
Inanimate flicker with animal vs.inanimate flicker w/out animal
p b 0.048 p b 0.974 p b 0.043 p b 0.866
7M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
newly created ‘no change’ flicker stimuli (i.e., the noise)were presentedalong with stimuli already used in experiment 1, where visual changesoccur (i.e., the signal).
3.1.1. Study hypothesesWe hypothesized that participants would be most sensitive to ani-
mals appearing and disappearing within a scene. The basis for this pre-diction lies with the notion of adaptive attention, suggesting thatgreater sensitivity to animate stimuli provides an advantage. If an atten-tional/perceptual bias for animacy exists then the animate signal will bestronger (high d′), increasing chances of attending to survival-relevantinformation in our environment. In contrast, the inanimate signal wasexpected to be less salient. Reduced sensitivity to inanimate objects(low d′)was anticipated as a reflection of an attentional/perceptual pro-cess that deprioritizes information that is less useful to our survival.
It was also predicted that the presence of animals in the scenewoulddramatically reduce a person's sensitivity (lower d′) to inanimatescene-changes.Wehypothesized that the presence of amore salient an-imate signal would serve as a distraction. The competing animate signalwas expected to reduce the attention/perceptual systems' capacity fordistinguishing a less-salient, inanimate signal from the noise.
Lastly, consistent with Haselton's error management theory(Haselton & Buss, 2009) a liberal response bias (low β) was anticipatedfor scenes where animals flickered.We expected that attention evolvedto help us adapt and survive would use a ‘better-safe-than-sorry’strategy when responding to animate stimuli. If the odds for survivalare improved by attending to changes in living things, there may beadded benefit from attentional strategies that are ‘liberal’ in how theyattend to animate changes. A mechanism prone to occasional errors,where people falsely over-respond as if an animate changes did occur(i.e., more false positive errors), would reduce the likelihood of missingsurvival-relevant information (i.e., more correct hits).
The most conservative response bias (high β) was anticipated inscenes where inanimate objects appeared and disappeared in the pres-ence of animate distractors. We expected that in the presence of a moresalient animate signal, the decision to focus on something inanimatewould require the most time and visual information.
3.2. Methods
3.2.1. ParticipantsParticipants for experiment 2 were recruited using the same means
as experiment 1. All were volunteer college students, 18-years or older,with normal or corrected-to-normal vision. No onewho took part in ex-periment 1 enrolled in experiment 2. A total of N=188 undergraduatevolunteers (116 F) were recruited with a mean age of 19.41 years.Trained research assistants were utilized to secure informed consentand debrief participants.
3.2.2. Materials and apparatusExperiment 2 was conducted using the same computer hardware
and software as experiment 1. Thirty of the 33 photograph scenesfrom experiment 1 were retained for use in experiment 2. Even with
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
the removal of 3 pictures, the equivalence of the scenes was retainedon key variables such as retinal size, luminance, contrast, and locationof the change.
Four versions of the 30 retained scenes were used. Like experiment1, one version of the photograph had the animate target removedfrom the scene, a second had the inanimate target removed from thescene, and a third removed both the animate and inanimate targets.The fourth variation of each photograph made no changes to the scene(i.e., both animate and inanimate objects were present).
Experiment 2 incorporated 90 of the 99 flicker task stimuli thatwereused for experiment 1. These served as the ‘signal’ stimuli, depictingvisual ‘change’ occurring in a scene. The nine unused stimuli werescene-pairs associated with the three excluded photographs(see Table 1). In total, experiment 2 used thirty ‘animal flicker: change’stimuli, thirty ‘inanimate flickerwith animal: change’ stimuli, and thirty‘inanimate flicker w/out animal: change’ stimuli.
Additionally, 90 ‘no change’ flicker task stimuli were produced forexperiment 2. These acted as the ‘noise’ stimuli, depicting quickswitches between the original photograph, a white screen mask, andthe same unchanged photograph. Thirty ‘animal flicker: no change’and thirty ‘inanimate flicker with animal: no change’ stimuli were pro-duced. Thirty ‘inanimate flicker w/out animal: no change’ stimuli werealso generated. These depicted a rapid alternation between a scenewithout the animal, a white mask, followed by the same photograph.
Six unique scene-presentation sets/lists were created to achieve acounterbalanced design (see Fig. 3). The goal was to show each partici-pant each of the 30 visual scenes. Yet, each scene could serve as a stim-ulus for any of the six conditions (e.g., animal flicker: no change, animalflicker: change, inanimate flicker with animal: no change, inanimateflicker with animal: change). Each presentation set ensured that all par-ticipants saw all of the 30 visual scenes in only one condition, withoutseeing any of the scenes more than once. Each set also guaranteedthat participants saw five ‘change’ and five ‘no change’ scenes-pairsfrom each of the stimulus-type category (e.g., animal flicker), and atotal of 15 signal (change) and 15 noise (no change) trials. Stimulus pre-sentation was randomized for every participant.
3.2.3. ProceduresParticipantswere seated approximately 61 cm in front of a computer
monitor, presented with on-screen instructions, given a chance to askquestions, and self-started the experimental flicker task. At the start ofthe experiment, participants were randomly assigned to one of the sixpresentation sets (see Fig. 3). They were told that their task was to de-termine if there was an object appearing and disappearing within thescene. Participants were instructed to press the “C” key to indicatethat they did detect a change. On the other hand, the “N” key-presswould indicate that no change was detected. Participants were askedto respond with the index fingers of their right and left hands. Partici-pants were specifically instructed to decide if changes were, or werenot occurring in the flickering scene-pairs. They were never told tolook for specific kinds of changes (i.e., animate vs. inanimate). Asingle-blind design kept participants naive to study hypotheses regard-ing the animate attentional bias.
eference for animacy impacts change detection, Evolution and Human
8 M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
Just like experiment 1, a black fixation cross on a white backgroundappeared in the center of the screen for 500ms prior to the presentationof each flicker. Timing of each scene-pair presentation (i.e., 250 ms foreach version of the scene) was kept consistent with experiment 1.Each flicker stimulus was displayed for 60 seconds, unless the partici-pant made a response.
Allowing participants to self-stop stimulus presentation maintainedsignal strength at threshold. Presenting only one flicker, for a period of1 second, would have resulted in responses based more on guessingthan signal detection. Flicker stimuli, within each presentation set,were presented in random order. Once 30 flicker stimuli had beenviewed by each participant, they were debriefed and excused.
In experiment 2, the variables of primary interest includedparticipants' sensitivity to the appearing/disappearing target stimuli(d′) and participants' response bias (β). Variables of secondary interestwere rates of hits, misses, correct rejections, and false alarms, as well astheir associated response speeds.
3.2.4. AnalysesIndependent, one-way analyses of variance (ANOVA) tests were
carried out to examine differences in participants' sensitivity (d′) andresponse bias (β) across the three stimulus-types. Like in experiment1, ANOVA tests used stimulus-type as the fixed independent factor[i.e., animal flicker, inanimate flicker with animal, and inanimate flickerw/out animal]. ANOVAs were followed by post-hoc Tukey HSDcomparisons. Sensitivity (d′) and response bias (β)means and standarddeviations were reported as (M± SD).
Participants' accuracy and response speed were also evaluated.Additional one-factor analysis of variance (ANOVA) tests, with post-hoc Tukey HSD analyses, were conducted to compare differences in hitand miss ratios. Associated response speed (RT), across the three‘change’ stimuli-types, were analyzed in the same manner. Speed andfrequency of correct rejections/false alarms, derived from responses tothe ‘no change’ flicker stimuli, were contrasted using independent-samples t-tests.
3.3. Results
3.3.1. Preliminary data aggregationResponses were coded as a ‘hit’, ‘miss’, ‘false alarm’, or ‘correct rejec-
tion’ (see Fig. 4). Subsequently, d′ and β were calculated according tostandards outlined in Stanislaw and Todorov (1999) and Abdi (2007).
As in experiment 1, all outliers were removed before conductingANOVA tests. Any hit, false alarm, or correct rejection response thatcorresponded to an RT b 500 ms was excluded. Any responses with RTsfalling outside the 95th percentile (i.e., standard Z-statistic = ±1.96)
Animal Flicker Inan
Presentation List1
Scenes 1-5: ChangeScenes 6-10: No Change
ScenScene
Presentation List2
Scenes 1-5: No ChangeScenes 6-10: Change
SceneScen
Presentation List3
Scenes 11-15 ChangeScenes 16-20 No Change
ScenScene
Presentation List4
Scenes 11-15 No ChangeScenes 16-20 Change
SceneScen
Presentation List5
Scenes 21-25 ChangeScenes 26-30 No Change
ScScene
Presentation List6
Scenes 21-25 No ChangeScenes 26-30 Change
ScenSce
Fig. 3. Stimulus content for 6 counter-balanced
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
were also removed. Approximately 7.1% of participants' original experi-ment 2 responses were withdrawn from further analysis (animal flicker:2.4%, inanimate flicker: 2.4%, and inanimate flicker w/out animal: 2.3%).
Using the remaining data, three aggregate d′ and β values werecalculated for each participant. Each composite index reflected aparticipant's average sensitivity and response bias for each of thethree stimulus-types. In total; five hundred fifty-five d′ and β values(n = 555) were created for evaluation with ANOVA. Nine participants'composites could not be computed due to removal of extreme scores.
Hits and miss rates were calculated using participants' responses tothe ‘change’ flicker stimuli-types. Correct rejection and false alarm ra-tios, based on participants' responses to ‘no change’ stimuli, were com-puted as additional measures of accuracy. Raw response times (RTs)associated with each accuracy index were log-transformed and usedin all ANOVA and t-test analyses. Raw RTs, associated with each accura-cy index, were log-transformed and used in all ANOVA and t-test anal-yses. Transformed RTs and SDs were reported as (RTlog/SDlog).
3.3.2. Principal analyses of beta and d-primeOne-way ANOVA results revealed statistically significant differences
in sensitivity (d′), when comparing participants' responses to the threevariations of stimuli (F2,552 = 64.02, p b 0.0001; η2 = 0.188). Partici-pants were most sensitive to changes in a scene when the appearing/disappearing targets were animals (d′ = 1.72 ± 0.57). Sensitivity wasreduced when participants had to notice the flickering of inanimate ob-jects, regardless of whether animals were visible (d′ = 1.03 ± 0.63) ornot visible (d′ = 1.18 ± 0.64).
ANOVA findings also yielded a significant main effect of stimulus-type (F2,552 = 7.16, p b 0.001; η2 = 0.025) for response bias (β). Amore conservative response style was associated with the detection ofinanimate objects (see Fig. 5). Participants responded to inanimate tar-gets when certainty of the target's appearance/disappearance was rela-tively high, yielding fewer hits but fewer false alarms. This was true incases where animals were present (β = 1.70 ± 0.53) and absent(β = 1.73 ± 0.55) from the scene. In contrast, participants were lessconservative when responding to animate stimuli (β = 1.53 ± 0.57).Less information was needed before responding to the perceived ap-pearance/disappearance of animate targets, resulting in more hits,fewer misses, but an increased rate of false alarms.
As outlined in Table 3, post-hoc test found differences in d′ and β,when comparing scenes with animate versus inanimate targets. TukeyHSD contrasts confirmed that participants were less sensitive andmore cautious when trying to identify inanimate scene-changes.
The presence of animal distractors impacted participants' sensitivity(d′) to the flickering of inanimate targets (two-tailed Tukey HSD:p b 0.043). Findings suggested that inanimate scene-changes were
imate Flicker withAnimal
Inanimate Flicker w/outAnimal
es 11-15: Changes 16-20: No Change
Scenes 21-25 ChangeScenes 26-30: No Change
s 11-15 No Changees 16-20 Change
Scenes 21-25 No ChangeScenes 26-30 Change
es 21-25 Changes 26-30 No Change
Scenes 1-5 ChangeScenes 6-10 No Change
s 21-25 No Changees 26-30 Change
Scenes 1-5 No ChangeScenes 6-10 Change
enes 1-5 Changes 6-10 No Change
Scenes 11-15 ChangeScenes 16-20 No Change
es 1-5 No Changenes 6-10 Change
Scenes 11-15 No ChangeScenes 16-20 Change
presentation stimuli lists for experiment 2.
eference for animacy impacts change detection, Evolution and Human
Fig. 4. Study 2 response coding, using signal detection theory.
9M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
significantly harder to notice when animals were also visible in thescene. Seeing versus not seeing an animal in the scene, while aninanimate object flickered, did not make a difference in participants'response style (two-tailed Tukey HSD for β, p b 0.866).
3.3.3. Secondary analyses of accuracy and response speedSummary statistics for experiment 2 accuracy and log-transformed
response rates are provided in Tables 4 and 5. One-wayANOVA revealedsignificant differences in hit/miss rates, when comparing participants'responses to the three variations of ‘change’ stimuli (F2,560 = 69.88,p b 0.0001; η2 = 0.202). Participants were most accurate when detect-ing animate changes (see Table 4). Comparatively, theywere least accu-rate when detecting inanimate changes in scenes with animals (two-tailed Tukey HSD; p b 0.0001) and without animals (two-tailed TukeyHSD; p b 0.0001) in them. A strong trend also suggested that inanimatechangeswere detectedwith greater accuracy in scenes that did not haveanimate distractors (two-tailed Tukey HSD; p b 0.064).
Parallels between experiment 1 and experiment 2 were also evidentwhen analyzing the response speeds (RTlog) linked to hits and misses.One-way ANOVA findings reflected significant differences in Hit RTlog(F2,530 = 34.14, p b 0.0001; η2 = 0.115) and miss RTlog (F2,476 = 3.19,p b 0.042; η2 = 0.013), between the three ‘change’ stimulus-types.
Participants were fastest when correctly identifying animate chang-es (see Table 4). Relatively, correctly identifying inanimate changes tooklonger, regardless of whether the scene depicted an animate distractor(two-tailed Tukey HSD; p b 0.0001) or not (two-tailed Tukey HSD;p b 0.0001). Misses were made significantly faster when an animatechange went undetected (see Table 4). This is compared to noticeably
Fig. 5. D-prime and beta differences betwe
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
slower miss rates in scenes where inanimate changes were occurring,with the animal absent from the scene (two-tailed Tukey HSD;p b 0.032)
Visual inspection of ‘no change’ summary data (see Table 5) sug-gested participants were more accurate when identifying the absenceof change in scenes without animate stimuli. Participants also seemedmore likely to make false alarm errors, when responding to ‘no change’scenes with animate distractors present.
T-test analyses, however, found no differences in correct rejection orfalse alarm rates between the two types of ‘no change’ scenes (t553 =1.33, p= 0.183, power= 0.2649). Therewere also no statistical distinc-tions in participants' correct rejection RTlog (t553 = − .596, p = 0.552,power = 0.0902) or false alarm RTlog (t130 = 0.009, p = 0. .993,power= 0.050), when comparing ‘no change’ scenes with and withoutanimate distractors. Lack of significant findings was likely due to lowstatistical power for detecting t-test effects.
3.4. Discussion
Results from experiment 2 indicated that animals were prioritizedby the human attention system. Participants were most sensitive and/or showed the greatest perceptual sensitivity (Morgan et al., 1990;Witt et al., 2015) toward changes in a scene when the appearing/disappearing targets were animals (higher d′). Animate changes weredetected with the greatest speed and degree of accuracy.
Participants demonstrated significantly decreased d-prime (d′)when trying to detect inanimate changes. Findings from studies like ex-periment 2, which utilize true signal-absent stimuli, let us conclude thatthe inanimate signal is perceived asweaker (Witt et al., 2015).Whetherbydecreased sensitivity or bymanifestation of an illusion-like perceptu-al bias (Morgan et al., 1990), the inanimate signal was perceived as lesssalient than the animate signal. Participants were slower and less accu-rate when detecting inanimate scene-changes.
The presence of animate distractors reduced participants' sensitivity,speed, and accuracy for detecting inanimate scene-changes. Higher d-prime (d′) was observed when participants had to identify inanimatechanges in scenes that did not depict animals. When animals werepresent and inanimate changehad to be identified, the salience of the an-imate stimuli reduced sensitivity to the inanimate change (lower d′).Even in ‘no change’ scenes, the presence of the animate distractors ap-peared to reduce correct rejection rates. It is presumed that participantscontinue to monitor the most-salient animate stimulus in a scene ham-pering their ability to detect presence or absence of inanimate change.
Analyses also reflected the use of a ‘better-safe-than-sorry’ decisionalprocess, when attending to animate information in the environment.Behavioral data showed absolutely no evidence of the behavioraldecision-based strategy known as the speed-accuracy tradeoff.
en the three experimental conditions.
eference for animacy impacts change detection, Evolution and Human
Table 4Descriptive data for the signal conditions in experiment 2.
Animacy condition Signal ‘change’ condition Mean hitpercentage (%)
SE of hitpercentage (%)
Mean hitRTlog (ms)
SE of hitRTlog (ms)
Mean misspercentage (%)
SE of Misspercentage (%)
Animal flicker Animal appears/disappears, inanimate objectvisible in the scene
74.21 1.45 3.64 0.01 25.79 1.44
Inanimate flicker w/out animal
Inanimate object appears/disappears, animalnot visible in the scene
50.65 2.01 3.78 0.02 49.35 2.01
Inanimate flickerwith animal
Inanimate object appears/disappears, animalvisible in the scene
44.71 2.08 3.83 0.02 55.29 2.08
10 M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
Participants were not slowing themselves down to improve accuracy,but were instead consistently faster and more accurate when identify-ing animate scene-changes.
In fact, participants demonstrated a more liberal behavioral re-sponse bias (low β) when they had to notice the flickering of animateobjects. Due to its design, experiment 2 produced metrics of beta (β)that could only be influenced by changes in the internal criterion(Witt et al., 2015). Thus, findings support the idea that a decisionalprocess plays a role in manifesting the observed animacy-selectivebias. Animate scene-changes were correctly identified fastest, furtherillustrating that less information was needed to produce a response toanimate scene-changes.
As anticipated, this over-response bias reduced participants' chancesofmissing any animate information. Increased accuracy in detecting an-imate changes seemed to accompany an increased tendency towardfalse alarm errors. Not only were participants more quick and accuratewhen responding to the animate-change signal, but animal stimuliagain appeared to interfere with inanimate scene-change detection.Participants were more likely to mistakenly think that scene-changeswere occurring in ‘no change’ scenes when irrelevant animalswere present.
Responding to inanimate changes was harder regardless of whetheranimals were visible. Participants' betas were indicative of a more con-servative response bias (high β). Inanimate changes were detectedmore slowly, suggesting that more information was required before in-animate changes triggered behavior. It took longest for participants tomiss inanimate changes in sceneswhere therewas no animate informa-tion. Inanimate change also was detected with less accuracy and withfewer false alarms. As such, a more inhibited response style was associ-ated with inanimate change-detection.
Taken together, it appears that the attention system is predisposedto de-prioritize and disregard inanimate information in our environ-ment, in favor of focusing on the animate. Broadly speaking, resultsfrom experiment 2 suggest that animacy influences sensitivity and thedetection strategies of participants.
4. General discussion
Our studies presented a methodologically novel exploration ofattentional preference for animacy. We also tested the hypothesis thatpresence of an animate stimulus would distract from detecting changesto inanimate objects in a scene. Previous studies of the animate moni-toring hypothesis test five categories of stimuli depicting two kinds of
Table 5Descriptive data for the noise conditions in experiment 2.
Animacy condition Noise ‘no change’ condition Mean CRpercenta
Animal or inanimateno-change with animal
Animate object is always visible,inanimate object also visible in the scene
94.63
Inanimate no-change w/out animal Inanimate object is always visible,animal not visible in the scene
96.11
Note: Correct rejections (i.e., correctly saying there is no change occurring within the scene), aFalse alarms (i.e., mistakenly saying there is no change occurring within the scene), are indicat
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
animate scene-changes and three kinds of inanimate scene changes(New et al., 2007, 2010). However, the presence of animate distractorsin those stimuli was not controlled for and was not consistent acrossthe five levels of the independent variable. Our studies used variationsof the same scenes to create a more balanced design, control for thepresence of animate distractors, and test the animate monitoring hy-pothesis more precisely. Using the SDT paradigm we quantifiably illus-trate how humans' attention systems vary perceptual sensitivity andselectively employ different perceptual/behavioral response biases (β)when responding to animate and inanimate object.
4.1. Detection of animate changes
Experiment 1 builds on the body of evidence showing that peoplearemore rapid and accuratewhen detecting changes to animate stimuli.Experiment 2 adds something new in empirically showing increasedsensitivity (d′) to animate stimuli. A more liberal response style towardanimacy was also indicated suggesting that it is easier to attend to theanimate. We suggest that higher sensitivity for animate changes occursbecause the animal has greater perceptual salience relative to other vi-sual information. Our perceptual/cognitive systems appear to needless information from the environment to allocate attentional resourcestowards animate things.
Error management theory (Haselton & Buss, 2009) presents aclear framework for understanding people's tendency to over-respond and reduce their chances of missing any animate informa-tion. This theory suggests that humans are biased towards deci-sions employing a better-safe-than-sorry strategy. When the costof false-negative errors is greater than the potential gain fromfalse-positive errors, such decisions likely improve survival and fa-cilitate decisiveness in the face of ambiguous information(Haselton & Buss, 2009).
Results from experiment 2 suggest participants used just such astrategy to allocate attentional resources. When animate and inanimateinformation was depicted in a scene, participants were more likely tomistakenly respond, indicating the presence of a flicker even thoughno flicker occurred. This false-alarm error was made less frequentlywhen ‘no change’ stimuli portrayed only inanimate information. Resultspoint to an attention system that is prone to occasionalmisallocations ofresources in the interest of reducing our chances of missing animatechanges in our environment. Changes that are particularly relevant toour survival (Tooby & Cosmides, 2005).
ge (%)SE of CRpercentage (%)
Mean CRRTlog (ms)
SE of CRRTlog (ms)
Mean FApercentage (%)
SE of FApercentage (%)
0.68 4.10 0.01 5.37 0.68
0.79 4.11 0.01 3.89 0.94
re indicated by (CR).ed by (FA).
eference for animacy impacts change detection, Evolution and Human
11M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
4.2. Do animate distractors interfere with detection of changes toinanimate stimuli?
In experiment 1, participants were significantly slower whenresponding to inanimate scene-changes when animals were also visiblein the scene. In addition though accuracy was not significantly influ-enced by the presence of animate distractors, inanimate objects werenoticed more readily when they were closer to animate non-targets.We presume this occurs because participants first attend to, and thenmonitor the animate stimulus while looking for the change. Changesoccurring close to the animal would naturally be noticed sooner thanchanges occurring further away.
Results from experiment 2 demonstrated that presence of animatedistractors impacted sensitivity to inanimate changes. Participantswere less sensitive to inanimate scene-changes when animals werepresent. The animate signal is most salient. Presence of animate infor-mation also clearly reduces the saliency of other inanimate informationin the scene.
4.3. Limitations
There are two alternative plausible explanations to our results. First,it can be argued that participants employed a ‘spot the animal first’strategy, and were consequently better at spotting animals. Second, itmay be that, despite being well camouflaged, animate changes ‘popout’ in a scene full of inanimate objects making them easier to noticethan inanimate changes among inanimate objects. Both interpretationssuggest that animate changes are noticed more readily because theyare visual ‘odd-balls’ in the scene. We agree that this is a limitation ofour studies.
Including a single animal in each scene was a deliberate choiceintended to test our secondary hypothesis about animate distractors.Substantial effort was taken to control for stimulus salience. Animatetargets werewell camouflaged. Theywere equated to inanimate targetsonmultiple perceptual features including color, size, luminance, andpo-sition in the scene. Additionally, when possible, inanimate targets werechosen specifically for their distinctiveness. Approximately 33% of ourinanimate were unique/singular trees, rocks, or bushes in the scene.
It is also worth noting that animate scene-changes comprised a mi-nority of total trials. In experiment 1, the animal flickered in 33% of trialswhile inanimate object flickered 66% of the time. In experiment 2,animate changes comprised a mere 16% of the total trials. Even ifsome participants initially sought out the animal first, they wouldhave soon realized that this strategy was inefficient for locating thechange across trials.
Lastly, it may be safe to discount alternative explanationsbecause our findings converge with previous studies of the animate-monitoring hypothesis. Animals receive preferential attention whenperceptual load is high (Calvillo & Jackson, 2014). Change-detection isfaster and more accurate for animate targets when inanimate targetsinclude tools, buildings, plants, or vehicles (New et al., 2007, 2010).The effect even persists when the inanimate targets have evolutionaryrelevance (e.g. tools, fruits) (Jackson & Calvillo, 2013). Taken together,it is difficult to imagine a more methodologically appropriate or un-tested inanimate counterweight to the animate targets in change-detection experiments that use natural scenes.
Future investigations might address these limitations by usingscenes with large numbers of animals and very few inanimate objects.For example, scenes containing large herds of animals on sparsely land-scaped grasslandswould facilitate testing of the ‘visual-oddball’hypoth-esis. These kinds of stimuli may make inanimate the features of a scenemore unique, less abundant, andmore distinct relative to the remainingcontent of the scene. Nevertheless, the animate monitoring hypothesiswould still predict that animate scene-changeswould receive attention-al priority over the inanimate objects. Future studies are already
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
planned to test the effects of increasing the number of animatedistractors in a scene.
5. Conclusion
Evolutionary psychologists suggest that attention should be concep-tualized as several domain-specific processes designed to solve particu-lar adaptive problems (Tooby & Cosmides, 2005). The animatemonitoring hypothesis predicts that selection tuned the human atten-tion system to prioritize and monitor animate stimuli in visual search(New et al., 2007). Results from our experiments lend additional sup-port to the animate monitoring hypothesis. Current findings suggestthat ongoing monitoring of animate stimuli in a visual scene distractsfrom attending to inanimate objects. Understanding the distractingquality of animals offers both real world implications and points to ad-ditional testable hypotheses. In busy traffic, a child safelywaiting on thesidewalk may draw more attention than a traffic signal turning red. Incontrolled research paradigms, increasing animate distractors in ascene should also increase the number of stimuli that must be moni-tored, leading to impaired task performance.
Appendix A. Equating experiment 1 visual stimuli on all relevantperceptual properties
All animals, in each picture, were camouflaged in their environment,and were similar in color and texture to the backgrounds of their corre-sponding scenes. None of the pictures contained animals silhouettedagainst the horizon. This was done to avoid “pop-out” effects of the an-imals, compared to other elements in the pictures. Researchers madesure the scenes represented a variety of natural environments, includ-ing deserts, grasslands, wooded forests, and mountain terrains. Imagescontaining people were excluded because New et al. (2010; 2007) al-ready found little difference in attentional bias for animals versus peo-ple. Excluding images of people avoided possible confounding ofanimate stimuli. Images depictingman-made artifacts (e.g. roads, build-ings, cars) were also omitted, focusing instead on ancestral- andsurvival-relevant stimuli. This was in keeping with our emphasis onhow attentionmay have adapted in response to naturally-occurring en-vironmental pressures. Absence of man-made artifacts also removedthe potential for a ‘familiarity’ confound.
Researchers then selected an animate and inanimate target withineach picture. Each picture was digitally manipulated to equate thesize, on average, of animate versus inanimate targets. Similar graphicaladjustments were performed to make animate and inanimate objectscomparable in their luminance when all stimuli form all pictureswere averaged.
Congruence of the targets was also controlled for, with the help ofthe volunteer standardization sample (n = 36). Researchers defined‘congruence’ as how well the object belonged, or “fit”, with the themeof the picture. Volunteers rated each pre-chosen animate and inanimatetarget for target-scene congruence, on a five-point Likert scale. Onlyscenes where targets received a rating of 4 (congruent with the scene)or 5 (very congruent with the scene) were retained.
Consideration was also given to ‘center of interest’ when selectingthe pictures. Researchers defined center of interest as the portion ofthe scene which initially draws the eyes, based on the composition ofthe picture. The volunteers in the standardization sample were askedto view each picture, with the pre-chosen animate and inanimate tar-gets removed from each scene, for 5000 ms. They were then shownthe same scene overlaid with a 4 × 4 grid and asked to indicate thequadrant (i.e., portion of the scene) that first drew their gaze. Agree-ment ranged from 85–100%. Researchers then adjusted and/or retainedpictures to ensure that animate and inanimate targets appeared withthe same frequency inside, and outside, the center of interest.
Location of animate and inanimate objects, in each scene, was alsocontrolled. Researchers divided each scene into four equidistant vertical
eference for animacy impacts change detection, Evolution and Human
12 M.N. Altman et al. / Evolution and Human Behavior xxx (2015) xxx–xxx
quadrants. If the target object fell within the two center quadrants it wascoded as “central”, and if an object fell either in the left or right quadrantit was coded as “peripheral”. Scenes were chosen so they would becounterbalanced, with 15 of each target type (i.e., animate andinanimate) falling in the center and periphery. Three retained sceneshad targets falling on the dividing lines between center and periphery.
To ensure that perceptual bottom-up featureswere equated asmuchas possible, researchers prepared two different flicker tasks designed toremove context from the scene and test for any remaining effects asso-ciated with stimulus salience. We applied a Gaussian blur function toeach scene and each image was rotated 180°. For each flicker, the firstmanipulated image was presented for 250 ms, followed by a whitemask, followed by a second manipulated picture with either theanimate or the inanimate stimulus removed for 250 ms, followed by asecond white mask. The flicker was repeated until participants pressedthe spacebar. Pictures were retained only if, for each scene, blurred an-imate and inanimate targets were detected at similar rates and withsimilar accuracy by the volunteers in the standardization sample. Thiscontrolled for top-down perceptual features, allowing us to test forany remaining bottom-up “pop-out” effects.
References
Abdi, H. (2007). Signal detection theory (SDT). In N. Salkind (Ed.), Encyclopedia of Mea-surement and Statistics (pp. 1–9). Thousand Oaks, CA: Sage.
Baron-Cohen, S. (1995).Mindblindness: An essay on autism and theory of mind. Cambridge,MA: MIT Press.
Barrett, H. C. (2005). Adaptations to Predators and Prey. In D. Buss (Ed.), The Handbook ofEvolutionary Psychology (pp. 200–223). Hoboken, NJ: Wiley & Sons.
Calvillo, D. P., & Jackson, R. E. (2014). Animacy, perceptual load, and inattentional blind-ness. Psychonomic Bulletin & Review, 21(3), 670–675. http://dx.doi.org/10.3758/s13423-013-0543-8.
Caramazza, A., & Shelton, J. R. (1998). Domain-specific knowledge systems in the brain:The animate–inanimate distinction. Journal of Cognitive Neuroscience, 10(1), 1–34.
Cavanagh, P., Labianca, A. T., & Thornton, I. M. (2001). Attention-based visual rou-tines: Sprites. Cognition, 80(1–2), 47–60. http://dx.doi.org/10.1016/S0010-0277(00)00153-0.
Cutting, J., & Kozlowski, L. (1977). Recognizing friends by their walk: Gait perceptionwithout familiarity cues. Bulletin of the Psychonomic Society, 9(5), 353–356. http://dx.doi.org/10.3758/BF03337021.
Opfer, J., & Gelman, S. (2011). Development of the animate–inanimate distinction. In U.Goswami (Ed.), The Wiley-Blackwell handbook of childhood cognitive development(pp. 213–238) (2nd ed.). Oxford, England: Wiley-Blackwell.
Hansen, C. H., & Hansen, R. D. (1988). Finding the face in the crowd: An anger superiorityeffect. Journal of Personality and Social Psychology, 54, 917–924.
Haselton, M., & Buss, D. M. (2009). Error management theory and the evolution ormisbeliefs. Behavioral and Brain Sciences, 32(6), 522–523.
Hodges, J. R., Graham, N., & Patterson, K. (1995). Charting the progression in seman-tic dementia: Implications for the organisation of semantic memory. Memory,3(3–4).
Hollingworth, A., & Henderson, J. M. (2000). Semantic informativeness mediates thedetection of changes in natural scenes. Visual Cognition, 7(1–3), 213–235 (Retrievedfrom http://www.tandfonline.com/doi/abs/10.1080/135062800394775).
IBM Corp (2013). IBM SPSS Statistics for Windows. Armonk, NY: IBM Corporation.Jackson, R. E., & Calvillo, D. P. (2013). Evolutionary relevance facilitates visual processing.
Evolutionary Psychology, 11(5), 1011–1026.
Please cite this article as: Altman, M.N., et al., Adaptive attention: How prBehavior (2015), http://dx.doi.org/10.1016/j.evolhumbehav.2016.01.006
Johnson, A., & Proctor, R. W. (2004). Information processing and the study of attention.Attention: Theory and practice (pp. 46–51). Thousand Oaks, CA: Sage Publications, Inc.
Lipp, O. V., Derakshan, N., Waters, A. M., & Logies, S. (2004). Snakes and cats in the flowerbed: Fast detection is not specific to pictures of fear-relevant animals. Emotion(Washington, D.C.), 4(3), 233–250. http://dx.doi.org/10.1037/1528-3542.4.3.233.
Lundqvist, D., & Ohman, A. (2005). Emotion regulates attention: The relation between fa-cial configurations, facial emotion, and visual attention. Visual Cognition, 12(1),51–84. http://dx.doi.org/10.1080/13506280444000085.
McAnally, K. I., Martin, R. L., Eramudugolla, R., Stuart, G. W., Irvine, D. R. F., & Mattingley, J.B. (2010). A dual-process account of auditory change detection. Journal ofExperimental Psychology: Human Perception and Performance, 36(4), 994–1004.http://dx.doi.org/10.1037/a0016895.
Morgan, M. J., Hole, G. J., & Glennerster, A. (1990). Biases and sensitivities in geometricalillusions. Vision Research, 30(11), 1793–1810. http://dx.doi.org/10.1016/0042-6989(90)90160-M.
New, J., Cosmides, L., & Tooby, J. (2007). Category-specific attention for animals reflectsancestral priorities, not expertise. Proceedings of the National Academy of Sciences,104(42), 16598–16603. http://dx.doi.org/10.1073/pnas.0703913104.
New, J., Schultz, R. T., Wolf, J., Niehaus, J. L., Klin, A., German, T. C., & Scholl, B. J. (2010). Thescope of social attention deficits in autism: Prioritized orienting to people and ani-mals in static natural scenes. Neuropsychologia, 48(1), 51–59. http://dx.doi.org/10.1016/j.neuropsychologia.2009.08.008.
Ohman, A., Flykt, A., & Esteves, F. (2001). Emotion drives attention: Detecting the snake inthe grass. Journal of Experimental Psychology: General, 130(3), 466–478. http://dx.doi.org/10.1037/AXJ96-3445.130.3.466.
Palmer, J., Verghese, P., & Pavel, M. (2000). The psychophysics of visual search. VisionResearch, 40, 1227–1268. http://dx.doi.org/10.1016/S0042-6989(99)00244-8.
Pastore, R. E., & Scheirer, C. J. (1974). Signal detection theory: Considerations for generalapplication81(12), 945–958.
Pavlova, M., Kragloh-Mann, I., Sokolov, A., & Birbaumer, N. (2001). Recognition of point-light biological motion displays by young children. Perception, 30, 925–933.
Pratt, J., Radulescu, P. V., Guo, R. M., & Abrams, R. a. (2010). It's alive! Animatemotion cap-tures visual attention. Psychological Science, 21, 1724–1730. http://dx.doi.org/10.1177/0956797610387440.
Psychology Tools Software Inc. (2012). E-Prime. Sharpsburg, PA: Psychology SoftwareTools Inc (Retrieved from http://www.pstnet.com).
Rensink, R. A., O'Regan, J. K., & Clark, J. J. (1997). To see or not to see: The need for atten-tion to perceive changes in scenes. Psychological Science, 8(5), 368–373 (Retrievedfrom http://pss.sagepub.com/content/8/5/368.abstract).
Simons, D., & Levin, D. (1997). Change blindness. Trends in Cognitive Sciences, 1(7),261–267.
Spelke, E., Phillips, A., & Woodward, A. (1995). Infants' knowledge of object motion andhuman action. In D. Sperber, D. Premack, & A. J. Premack (Eds.), Causal cognition: Amultidisiplinary debate (pp. 44–78). New York: Clarendon Press.
Stanislaw, H., & Todorov, N. (1999). Calculation of signal detection theory measures.Behavior ResearchMethods, Instruments, & Computers: A Journal of the Psychonomic So-ciety, Inc., 31(1), 137–149 (http://www.ncbi.nlm.nih.gov/pubmed/10495845).
Sui, J., & Liu, C. H. (2009). Can beauty be ignored? Effects of facial attractiveness on covertattention. Psychonomic Bulletin & Review, 16(2), 276–281. http://dx.doi.org/10.3758/PBR.16.2.276.
Tooby, J., & Cosmides, L. (2005). Conceptual foundations of evolutionary psychology. In D.Buss (Ed.), The Handbook of Evolutionary Psychology (pp. 5–67). Hoboken, NJ: JohnWiley & Sons Inc. http://dx.doi.org/10.3758/s13423-013-0543-8.
Verghese, P. (2001). Visual search and attention: A signal detection theory approach.Neuron, 31(4), 523–535. http://dx.doi.org/10.1016/S0896-6273(01)00392-0.
Werner, S., & Thies, B. (2000). Is “change blindness” attenuated by domain-specific exper-tise? An expert–novices comparison of change detection in football images. VisualCognition, 7(1–3), 163–173. http://dx.doi.org/10.1080/135062800394748.
Wilken, P., & Ma, W. J. (2004). A detection theory account of change detection. Journal ofVision, 4(12), 11-11. http://dx.doi.org/10.1167/4.12.11.
Witt, J. K., Taylor, J. E. T., Sugovic, M., & Wixted, J. T. (2015). Signal detection measurescannot distinguish perceptual biases from response biases. Perception. http://dx.doi.org/10.1068/p7908.
eference for animacy impacts change detection, Evolution and Human
